Package elki.clustering.kmeans.initialization

Class SphericalAFKMC2

  • All Implemented Interfaces:
    KMeansInitialization
    @Title("Spherical AFK-MC\u00b2")
@Reference(authors="R. Pratap, A. A. Deshmukh, P. Nair, T. Dutt",
           title="A Faster Sampling Algorithm for Spherical k-means",
           booktitle="Proc. 10th Asian Conference on Machine Learning, ACML",
           url="http://proceedings.mlr.press/v95/pratap18a.html",
           bibkey="DBLP:conf/acml/PratapDND18")
public class SphericalAFKMC2
extends AFKMC2
    Spherical K-Means++ initialization with markov chains.

    FIXME: currently assumes the vectors to be L2 normalized beforehand, but does not ensure that this is true.

    Reference:

    R. Pratap, A. A. Deshmukh, P. Nair, T. Dutt
    A Faster Sampling Algorithm for Spherical k-means
    Proc. 10th Asian Conference on Machine Learning, ACML

    Since:
    0.8.0
    Author:
    Erich Schubert

    • Nested Class Summary

      Nested Classes 
      Modifier and Type Class Description
      protected static class  SphericalAFKMC2.Instance
      Abstract instance implementing the weight handling.
      static class  SphericalAFKMC2.Par
      Parameterization class.

    • Field Summary

      Fields 
      Modifier and Type Field Description
      protected double alpha
      Parameter to balance distance vs. uniform sampling.
      private static elki.logging.Logging LOG
      Class logger.

      • Fields inherited from class elki.clustering.kmeans.initialization.KMC2

        m

    • Constructor Summary

      Constructors 
      Constructor Description
      SphericalAFKMC2​(int m, double alpha, elki.utilities.random.RandomFactory rnd)
      Constructor.

    • Method Summary

      All Methods Instance Methods Concrete Methods 
      Modifier and Type Method Description
      double[][] chooseInitialMeans​(elki.database.relation.Relation<? extends elki.data.NumberVector> relation, int k, elki.distance.NumberVectorDistance<?> distance)
      Choose initial means

      • Methods inherited from class java.lang.Object

        clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

    • Field Detail

      • LOG

        private static final elki.logging.Logging LOG
        Class logger.

      • alpha

        protected double alpha
        Parameter to balance distance vs. uniform sampling.

    • Constructor Detail

      • SphericalAFKMC2

        public SphericalAFKMC2​(int m,
                       double alpha,
                       elki.utilities.random.RandomFactory rnd)
        Constructor.
        Parameters:
        m - M parameter
        alpha - alpha-SKM parameter, usually 1.5
        rnd - Random generator.

    • Method Detail

      • chooseInitialMeans

        public double[][] chooseInitialMeans​(elki.database.relation.Relation<? extends elki.data.NumberVector> relation,
                                     int k,
                                     elki.distance.NumberVectorDistance<?> distance)
        Description copied from interface: KMeansInitialization
        Choose initial means
        Specified by:
        chooseInitialMeans in interface KMeansInitialization
        Overrides:
        chooseInitialMeans in class AFKMC2
        Parameters:
        relation - Relation
        k - Parameter k
        distance - Distance function
        Returns:
        List of chosen means for k-means